No AI Slop manifesto draws 442 signatures for human review of AI work

The people who send code or documents written by AI agents should disclose how the material was produced and check the parts their colleagues will rely on, according to a manifesto from speakers and attendees at AI Engineer Code Summit 2026. A Nov. 15 revision of the No AI Slop manifesto on GitHub Gist listed 442 signatures at 10:33 a.m. UTC.

Dexter Horthy, who introduced the manifesto at the San Francisco conference on Nov. 12, described implementation as one stage in a software team’s work. People still decide what to change, check whether the result meets the goal and look for feedback after release. For a pull request, the proposal calls for the person asking colleagues to merge a change to take responsibility for what those colleagues are being asked to review.

When a fast draft leaves a slow review

The manifesto uses slop for work that asks more effort of the reader than the sender put into producing it. A coding agent can quickly draft a long change. Understanding its purpose and how it fits the rest of a project still takes time. The signers apply that concern to plans and internal messages as well as code.

Horthy’s talk addressed the steps around the agent. He said teams need ways both to start useful work and to make the result meet a standard. A team can feed test results and automated checks back to the agent so that it corrects some problems before a reviewer gets involved. Horthy presented this as a way to organize the work. He gave no measured change in review time for that arrangement.

Which checks to automate

The signers propose scaling human review to the consequences of a change. A reviewer might inspect part of the code, a design decision or evidence about the result. For recurring mistakes they call for improving the system, such as by adding a repeatable rule to a linter, an automated tool that checks code for defined problems.

Horthy contrasted a small interface change and a decision about database structure. In the first case a preview can help resolve a question about appearance. With a schema change, he said, it makes sense to consider earlier how expensive it would be to redo the work. A check has to test something that could actually go wrong with that change.

The talk also made room for experiments kept behind a feature flag, a setting that can leave unfinished work hidden from users. Horthy argued that once a team needs to build on a change, the reasons for earlier decisions should be recoverable. A diff shows what changed. The context behind it helps the next person understand why.

The manifesto cites the contribution policy of Ghostty, an open-source terminal application. That policy asks outside contributors to disclose which AI tools they used and how much those tools helped. Contributors must also be able to explain how their changes work with the surrounding system. Ghostty uses that rule in its own project. The manifesto offers a framework for discussing where to set review boundaries across teams.

The Hacker News thread for the manifesto showed 220 points and 131 comments at about 11:02 a.m. UTC on Nov. 15. Commenters debated whether splitting large pull requests makes failures easier to isolate or gets in the way of larger rewrites. Those figures describe activity on the discussion thread.

The organizers collect signatures in comments on a GitHub Gist, a page for sharing a short document, and say they periodically add them to the list. GitHub reported 446 comments at about 11:04 a.m. UTC. The 10:33 revision of the list showed 442 signatures, up from 439 in the preceding revision. The manifesto’s separate website showed 352 entries when checked at 11:02. The three figures come from separate displays and count different things.